Human Pancreatic β Cell lncRNAs Control Cell-Specific Regulatory Networks

Article

Reference

Human Pancreatic β Cell lncRNAs Control Cell-Specific Regulatory Networks

AKERMAN, Ildem, et al.

Abstract

Recent studies have uncovered thousands of long non-coding RNAs (lncRNAs) in human pancreatic β cells. β cell lncRNAs are often cell type specific and exhibit dynamic regulation during differentiation or upon changing glucose concentrations. Although these features hint at a role of lncRNAs in β cell gene regulation and diabetes, the function of β cell lncRNAs remains largely unknown. In this study, we investigated the function of β cell-specific lncRNAs and transcription factors using transcript knockdowns and co-expression network analysis.

This revealed lncRNAs that function in concert with transcription factors to regulate β cell-specific transcriptional networks. We further demonstrate that the lncRNA PLUTO affects local 3D chromatin structure and transcription of PDX1, encoding a key β cell transcription factor, and that both PLUTO and PDX1 are downregulated in islets from donors with type 2 diabetes or impaired glucose tolerance. These results implicate lncRNAs in the regulation of β cell-specific transcription factor networks.

AKERMAN, Ildem, et al . Human Pancreatic β Cell lncRNAs Control Cell-Specific Regulatory Networks. Cell Metabolism , 2017, vol. 25, no. 2, p. 400-411

DOI : 10.1016/j.cmet.2016.11.016 PMID : 28041957

Available at:

http://archive-ouverte.unige.ch/unige:93706

Disclaimer: layout of this document may differ from the published version.

1 / 1

Article Human Pancreatic b Cell lncRNAs Control Cell-

Specific Regulatory Networks

Graphical Abstract

Highlights

d

A loss-of-function screen reveals functional

cell lncRNAs

d

Cell-specific lncRNAs and transcription factors regulate common gene networks

d

The lncRNA

influences interactions between an enhancer cluster and

d PLUTO

and

are deregulated in type 2 diabetes and impaired glucose tolerance

Authors

Ildem Akerman, Zhidong Tu, Anthony Beucher, ..., Eric Schadt, Philippe Ravassard, Jorge Ferrer

Correspondence

jferrerm@imperial.ac.uk

In Brief

Akerman et al. studied the function of human

cell lncRNAs with RNAi, CRISPRi, and co-expression networks.

This revealed

cell lncRNAs and

transcription factors that control common regulatory networks. One lncRNA,

is downregulated in type 2 diabetes and controls

key

cell transcription factor.

Accession Numbers

GSE83619

Akerman et al., 2017, Cell Metabolism25, 400–411 February 7, 2017ª2016 Published by Elsevier Inc.

http://dx.doi.org/10.1016/j.cmet.2016.11.016

Cell Metabolism

Article

Human Pancreatic b Cell lncRNAs Control Cell-Specific Regulatory Networks

Ildem Akerman,^1,2,3Zhidong Tu,⁴Anthony Beucher,¹Delphine M.Y. Rolando,¹Claire Sauty-Colace,⁵Marion Benazra,⁵ Nikolina Nakic,¹Jialiang Yang,⁴Huan Wang,⁴Lorenzo Pasquali,^3,6Ignasi Moran,¹Javier Garcia-Hurtado,^2,3

Natalia Castro,^2,3Roser Gonzalez-Franco,¹Andrew F. Stewart,⁷Caroline Bonner,⁸Lorenzo Piemonti,⁹Thierry Berney,¹⁰ Leif Groop,¹¹Julie Kerr-Conte,⁸Francois Pattou,⁸Carmen Argmann,⁴Eric Schadt,⁴Philippe Ravassard,⁵

and Jorge Ferrer^1,2,3,12,*

1Section of Epigenomics and Disease, Department of Medicine, Imperial College London, London W12 0NN, United Kingdom

2Genomic Programming of Beta Cells Laboratory, Institut d’Investigacions Biomediques August Pi I Sunyer (IDIBAPS), Barcelona 08036, Spain

3Centro de Investigacio´n Biome´dica en Red de Diabetes y Enfermedades Metabo´licas Asociadas (CIBERDEM), Madrid 28029, Spain

4Department of Genetics and Genomic Science, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA

5Sorbonne Universite´s, UPMC Univ Paris 06, INSERM, CNRS, Institut du cerveau et de la moelle (ICM) – Hoˆpital Pitie´-Salpeˆtrie`re, Boulevard de l’Hoˆpital, Paris 75013, France

6Germans Trias i Pujol University Hospital and Research Institute and Josep Carreras Leukaemia Research Institute, Badalona 08916, Spain

7Diabetes, Obesity, and Metabolism Institute, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA

8European Genomic Institute for Diabetes, INSERM UMR 1190, Lille 59800, France

9Diabetes Research Institute (HSR-DRI), San Raffaele Scientific Institute, Milano 20132, Italy

10Cell Isolation and Transplantation Center, University of Geneva, 1211 Geneva 4, Switzerland

11Department of Clinical Sciences, Lund University Diabetes Centre, Lund University, Lund 20502, Sweden

12Lead Contact

*Correspondence:jferrerm@imperial.ac.uk http://dx.doi.org/10.1016/j.cmet.2016.11.016

SUMMARY

Recent studies have uncovered thousands of long non-coding RNAs (lncRNAs) in human pancreatic

cells.

cell lncRNAs are often cell type specific and exhibit dynamic regulation during differentiation or upon changing glucose concentrations. Although these features hint at a role of lncRNAs in

cell gene regulation and diabetes, the function of

cell lncRNAs remains largely unknown. In this study, we investigated the function of

cell-specific lncRNAs and transcription factors using transcript knock- downs and co-expression network analysis. This revealed lncRNAs that function in concert with tran- scription factors to regulate

cell-specific transcrip- tional networks. We further demonstrate that the lncRNA

affects local 3D chromatin structure and transcription of

cell tran- scription factor, and that both

and

are downregulated in islets from donors with type 2 dia- betes or impaired glucose tolerance. These results implicate lncRNAs in the regulation of

cell-specific transcription factor networks.

INTRODUCTION

Transcriptome surveys have uncovered tens of thousands of mammalian transcripts longer than 200 nucleotides that have low protein-coding potential (Carninci et al., 2005; Derrien

et al., 2012; Guttman et al., 2009). A small fraction of these long non-coding RNAs (lncRNAs) have been shown to control gene expression by modulating chromosomal structure, tran- scription, splicing, mRNA transport, stability, or translation (Car- rieri et al., 2012; Chen and Carmichael, 2009; Gong and Maquat, 2011; Lai et al., 2013; Luco and Misteli, 2011; Willingham et al., 2005; Yao et al., 2010). Specific lncRNAs have thus been impli- cated in various key processes, including random X chromo- some inactivation, imprinting, the cell cycle, organogenesis, differentiation, pluripotency, and cancer progression (Guttman et al., 2011; Huarte et al., 2010; Hung et al., 2011; Klattenhoff et al., 2013; Kretz et al., 2013; Penny et al., 1996; Schmitt and Chang, 2013; Sleutels et al., 2002; Ulitsky et al., 2011). Despite these wide-ranging biological roles, the fraction of lncRNAs that is genuinely functional and the true impact of lncRNAs in hu- man biology and disease remain poorly understood.

Pancreaticbcells regulate glucose homeostasis by secreting insulin and play a central role in the pathogenesis of major forms of diabetes mellitus. Recently, more than 1,100 lncRNAs were identified in human pancreatic islets and purifiedbcells (Mora´n et al., 2012) as well as in mouse pancreatic islet cells (Benner et al., 2014; Ku et al., 2012; Mora´n et al., 2012). A large fraction of humanbcell lncRNAs are cell-specific, and several are known to be activated duringbcell differentiation (Mora´n et al., 2012).

This cellular specificity has also been noted for lncRNAs in other cell types (Cabili et al., 2011; Derrien et al., 2012) and points to the possibility that lncRNAs may regulate genetic programs important for lineage-specific differentiation or specialized cellular functions. Further, several bcell lncRNAs were shown to be regulated by extracellular glucose concentrations, sug- gesting a potential role of lncRNAs in the functional adaptation ofbcells to increased insulin secretory demands (Mora´n et al., 400 Cell Metabolism25, 400–411, February 7, 2017ª2016 The Authors. Published by Elsevier Inc.

This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

2012). Some islet lncRNAs map to loci that contain polygenic or Mendelian defects associated with human diabetes, whereas selected lncRNAs show deregulation in islets from organ donors with human type 2 diabetes (T2D) (Fadista et al., 2014; Mora´n et al., 2012). Collectively, these properties define a newly identi- fied class of candidate regulators ofbcell differentiation and function, with potential implications for human diabetes mellitus.

However, the true relevance of b cell lncRNAs depends on whether they elicit a physiological function in humanb cells, which remains to be addressed systematically.

In the current study, we have focused on a set of lncRNAs that show restricted expression in human pancreaticbcells and have tested the hypothesis that they regulatebcell gene expression.

Our studies have uncovered a regulatory network in which line- age-specific lncRNAs and transcription factors (TFs) control common genes. Furthermore, we show that lncRNAs frequently regulate genes associated with clusters of islet enhancers, which have previously been shown to be the primary functional targets of islet-specific TFs. We performed a detailed analysis of a specific lncRNA namedPLUTO, which controlsPDX1, a master regulator of pancreas development andbcell differenti- ation and, thereby, modulates thePDX1-dependent transcrip- tional program. Finally, we show that PLUTO and PDX1 are downregulated in islets from organ donors with type 2 diabetes or impaired glucose tolerance, suggesting a potential role in hu- man diabetes.

RESULTS

HumanbCell lncRNA Knockdowns Cause Profound Transcriptional Phenotypes

To directly test the regulatory function of pancreatic b cell lncRNAs, we carried out loss-of-function experiments in a glucose-responsive human isletbcell line, EndoC-bH1 (Ravas- sard et al., 2011). We chose a human model because only some human lncRNAs are evolutionary conserved (Derrien et al., 2012; Mora´n et al., 2012; Okazaki et al., 2002; Pang et al., 2006), and we perturbed the function of lncRNAs through RNAi-based transcript knockdowns rather than genomic dele- tions because deletions could potentially disruptcis-regulatory elements. We thus designed lentiviral vectors that contain RNA polymerase II-transcribed artificial microRNAs (hereafter referred to as amiRNA) with perfect homology to the target sequence to elicit target cleavage. The amiRNAs contain an arti- ficial stem sequence targeting our lncRNA of choice as well as flanking and loop sequences from an endogenous miRNA to allow their processing as pre-miRNA by the RNAi pathway (Fig- ure S1A). As a reference, we used the same strategy to knock down TFs that are well known to regulate gene expression in pancreatic islets as well as five different non-targeting amiRNA sequences as controls.

The lncRNAs selected for knockdown were derived from a short list of 25 lncRNAs that showed (1) a markedly enriched expression in human islets and fluorescence-activated cell sort- ing (FACS)-purifiedbcells relative to the exocrine pancreas and a panel of non-pancreatic tissues, (2) expression in the EndoC- bH1bcell line, and (3) a chromatin profile in human islets that was consistent with an active promoter (Figures S1C and S1D). Of these 25 lncRNAs, 12 were shortlisted because they

were near a protein-coding gene that has an important function inbcells. The lncRNAs had variable subcellular enrichment pat- terns (Figure S1B), and eight of the 12 lncRNAs had detectable transcripts in orthologous or syntenic mouse regions (Table S1;

Mora´n et al., 2012). We then screened four amiRNA sequences for each of the 12 lncRNAs and identified two efficient (>50%

knockdown) amiRNAs for seven lncRNAs and one efficient amiRNA sequence for the other five lncRNAs (Figure S1E). Two efficient amiRNAs were also obtained for five essential islet TFs (HNF1A,GLIS3,MAFB,NKX2.2, andPDX1).We thus trans- duced EndoC-bH1 cells with lentiviruses expressing each amiRNA. This was done in duplicate or in triplicate for lncRNAs that only had one efficient amiRNA. 80 hr post-transduction, RNA was harvested and hybridized to oligonucleotide microar- rays (Figure 1A). For each target gene, we combined expression data from all knockdowns and compared them to the control transductions with five different control amiRNAs to identify genes that were differentially expressed at a significance level of p < 10³(ANOVA) (Figure 1B).

As expected, knockdown of islet TFs consistently produced transcriptional phenotypes (Figure 1B). Remarkably, knockdown of 9 of the 12 islet lncRNAs also caused transcriptional changes (Figure 1B;Figure S1F). A more detailed analysis showed that some of the lncRNAs that presented knockdown phenotypes had visible effects on a neighboring gene, suggesting a possible cis-regulatory mechanism, although other such lncRNAs did not appear to affect neighboring genes and may thus function throughtrans-regulatory mechanisms (Figure 1E;Figure S1G).

These loss-of-function experiments with selected lncRNAs therefore suggested that lncRNAs can regulate the expression of pancreaticbcell genes.

Gene silencing using the RNAi pathway can theoretically lead to nonspecific gene deregulation. In our experimental model, a significant nonspecific result would occur when two unrelated amiRNAs elicited changes in a common set of genes that were not observed in the panel of control non-targeting amiRNAs.

To assess the likelihood that two unrelated amiRNA sequences elicit such an effect, we studied the five sets of control (non-tar- geting) amiRNAs, compared all ten possible combinations of two versus three control amiRNAs, and determined the number of differentially expressed genes (Figure 1C). Likewise, for each TF or lncRNA that had two valid amiRNAs, we compared the two target-specific amiRNAs against all possible combinations of three control amiRNAs (Figure 1C). As seen inFigure 1D, con- trol versus control comparisons generated a median of 16 (IQR = 15–22) differentially expressed genes, whereas all five TFs and six of the seven lncRNA knockdowns led to a significantly higher number of differentially expressed genes (Mann-Whitney test, p < 10⁴for all lncRNA/TF versus control comparisons except HI-LNC75, p = 0.004, and HI-LNC76, p > 0.5). These results show that the observed phenotypes are unlikely to be caused by unspecific effects of amiRNAs and indicate that the sequence-specific inhibition of selected islet lncRNAs can result in transcriptional changes comparable in magnitude to the inhi- bition of well-established islet transcriptional regulators.

The primary function ofbcells is to synthesize and secrete insulin in response to changes in glucose concentrations.

Among the genes that showed functional dependence on lncRNAs, we identified numerous genes that are known to

regulate transcription or secretion inb cells, including RFX6, PDX1,CACNA1D,ATP2A3,ROBO1and2,PDE8A,ATP6AP1, KCNJ15,TRPM3,ERO1LB, andHADH (Figure 2A; Anderson et al., 2011; Li et al., 2010; Louagie et al., 2008; Okamoto et al., 2012; Smith et al., 2010; Tian et al., 2012; Varadi and Rut- ter, 2002; Wagner et al., 2008; Yang et al., 2013; Zito et al., 2010).

We therefore measured insulin content and glucose-stimulated insulin secretion (GSIS) in T antigen-excised EndoC-bH3 cells after knocking down four lncRNAs that showed the strongest transcriptional phenotypes (HI-LNC12, HI-LNC78, HI-LNC80, andHI-LNC71). Congruent with the broad transcriptional pheno- type, we observed reduced insulin content and, consequently, Lentiviral transduction into

ENDOC-

RNA harvest (80 hrs) Affymetrix HTA2.0 analysis Validated amiRNAs:

- non-targeting controls (n=5, x2) - TFs (n=5, two amiRNA/target, x2) - lncRNAs (n=7, two amiRNA/target, x2) - lncRNAs (n=5, one amiRNAi/target,x3)

-1000 -500 0 500

1000 Upregulated Downregulated

HI-LNC12 HI-LNC15 HI-LNC30 HI-LNC75 HI-LNC76 HI-LNC78 HI-LNC80

GLIS3 HNF1A MAFB NKX2.2 PDX1 HI-LNC25 HI-LNC79 HI-LNC71 HI-LNC85 HI-LNC70

Number of differentially expressed genes A

Control vs. control

vs.

C3 C2

C1 C5 C4

Combination 1

Combination 2

Combination 10

...

vs.

C4 C2

C1 C5 C3

vs.

C5 C4

C3 C2 C1

Control vs TF or HI-LNC

Number of differentially expressed genes HI-LNC12

C

B

V1 V2 V3 V4 V5 V6 V7 V8 V9 V10 V11 V12 V13

0100200300400500600

0 300

***

***

***

*** ***

***

***

***

***

***

**

HI-LNC12 HI-LNC15 HI-LNC30 HI-LNC75 HI-LNC76 HI-LNC78 HI-LNC80

GLIS3 HNF1A MAFB NKX2.2 PDX1

CtlvsCtl

600

Differentially expressed genes from 10 combinations

ns D

E

0 0.4 0.8 1.2

UNC5A mRNA

Control amiRNA HI-LNC12amiRNA-1 HI-LNC12 amiRNA-2 0

0.4 0.8 1.2

NKX2.2 mRNA

*

Control amiRNA HI-LNC15amiRNA-1 HI-LNC15 amiRNA-2

* **

vs.

C3 C2 C1 Combination 1

Combination 2

Combination 10

...

vs.

C4 C2 C1

vs.

C5 C4 C3

sh1 sh2

sh1 sh2

sh1 sh2

Figure 1. Knockdown of SelectedbCell lncRNAs Leads to Transcriptional Phenotypes

(A) Schematic of the experimental plan. Lentivirally encoded amiRNAs were validated and transduced in duplicate (32) or triplicate (33) into ENDOC-bH1 cells as indicated and then analyzed with oligonucleotide expression arrays.

(B) Differential gene expression analysis revealed genes that show significant up- or downregulation after knockdown of TFs or lncRNAs. For each TF or lncRNA, we combined all replicates transduced with the different target-specific amiRNAs and compared these with all replicates from five non-targeting controls.

Differential expression was determined at p < 10³(ANOVA).

(C) We compared gene expression data from all ten possible combinations of three versus two control non-targeting amiRNAs. Similarly, the two independent amiRNAs that target each TF or lncRNA were compared with all ten possible combinations of three control amiRNAs. For this analysis, we only considered the seven lncRNAs that were targeted by two independent amiRNAs.

(D) Control comparisons result in a low number of differentially regulated genes (average 15 genes), whereas most TF and lncRNA comparisons yield higher numbers of differentially regulated genes. ***p < 10⁴; **p < 0.01; ns, not significant compared with control comparisons; Mann-Whitney test.

(E)HI-LNC15regulates its neighboring gene,NKX2.2, whereasHI-LNC12knockdown (KD) does not affect its adjacent active gene,UNC5A(left). Further ex- amples are shown inFigure S1G. RNAs were normalized toTBPmRNA and expressed relative to control amiRNAs; n = 3, error bars represent SEM; **p < 0.01,

*p < 0.05 (Student’s t test).

402 Cell Metabolism25, 400–411, February 7, 2017

impaired glucose-stimulated insulin secretion forHI-LNC12,HI- LNC78, andHI-LNC71knockdowns (Figure 2B). ForHI-LNC78, a glucose-regulated islet transcript (Mora´n et al., 2012) that is or- thologous to mouseTunarand zebrafishmegamind (linc-birc6) lncRNAs (Ulitsky et al., 2011), there was a reduction in GSIS after correcting for the reduction in insulin content (p = 0.002) (Fig- ure S2A). To further validate these effects, the same lncRNAs were downregulated using antisense locked nucleic acid (LNA GapmeRs, Exiqon) GapmeRs, which also led to impaired insulin secretion after knockdown of HI-LNC12 and HI-LNC78 (Fig- ure S2B). Taken together, lncRNA knockdown studies identified lncRNAs that modulate gene expression and, consequently, in- sulin secretion in a humanbcell line.

Human Islet lncRNAs and TFs Regulate Common Gene Expression Programs

To gain insight into the expression programs that are regulated by islet-specific lncRNAs and TFs, we compared their knock- down gene expression phenotypes. We first assessed changes in gene expression occurring after knockdown of the different islet TFs and found high Pearson correlation values for all pair- wise comparisons (r = 0.4–0.8, p < 10²⁷) (Figure 3A;Figure S3).

This finding is consistent with the notion that islet-specific TFs often bind to common genomic targets and function in a combi- natorial manner (Pasquali et al., 2014; Qiu et al., 2002; Wilson et al., 2003). Interestingly, the transcriptional changes that occurred after the inhibition of several lncRNAs significantly

ADCY8 COG3 COPG2 CTNNB1 DOPEY1 EXOC4 HADH KCNJ3 PDE8A ADCY8

ATP2A3 ATP6AP1 CACNA1A CACNA1D CADM1 CADPS CREBBP GNAS HADH

HI-LNC12

-cell function genes regulated by lncRNAs

ERO1LB HADH KCNJ3 TM4SF4 PDX1 VAMP3 HI-LNC78 HI-LNC71 KCNJ15

NFAT5 PAX6 PCSK2 PDE8A ROBO1 ROBO2 SCIN TM4SF4 TRPM3

PGK1 PRKAR2A RFX3 RFX6 ROBO1 SLC25A6 STAT3 TM4SF4 TMED10

A

0 10 20 30

2.8 mM 15 mM 2.8 mM +IBMX 15 mM +IBMX

**

***

**

Secretedinsulin (foldovercontrol)

0 5 10 15 20

2.8 mM 15 mM 2.8 mM +IBMX 15 mM +IBMX

*

** ₀ ⁵

10 15

2.8 mM 15 mM 2.8 mM +IBMX 15 mM +IBMX

*

* ** **

Control KD HI-LNC12 KD

*** ***

*** ** ** *

0 0.5 1

1.5 * * *

Insulincontent (foldovercontrol)

Control KD HI-LNC78 KD

Control KD HI-LNC71 KD

0 0.5 1 1.5

2.8 mM 15 mM

0 0.5 1 1.5

2.8 mM 15 mM

2.8mM 15mM 2.8mM 15mM

+IBMX

2.8mM 15mM 2.8mM 15mM

+IBMX

2.8mM 15mM 2.8mM 15mM

+IBMX

B

Figure 2. Knockdown of lncRNAs Impairs Insulin Secretion

(A) Examples of genes known to play a role inbcell function regulated by islet lncRNAs.

(B) Glucose-stimulated insulin secretion was tested on T antigen-excised EndoC-bH3 cells after transduction with amiRNAs targeting the indicated lncRNAs or controls. Secreted or total insulin content was normalized to the number of cells per well and expressed as fold change over control amiRNA treatment at 2.8 mM glucose. Each bar represents an average from two independent amiRNA vectors and 12 separate wells from two independent experiments. Error bars represent SEM; ***p < 10³, **p < 0.01, *p < 0.05 (Student’s t test).

correlated with those observed following inhibition of TFs (Figure 3A; Figure S3;

see also a cluster analysis of TF- and lncRNA-dependent changes inFigure 3B).

Some pairwise comparisons that illustrate this finding includeHI-LNC78-dependent gene expression changes, which corre- lated highly with HNF1A- and MAFB- dependent changes (Pearson’s r = 0.87 and 0.89, respectively, p < 10⁷¹), and HI-LNC15-dependent changes, which correlated with those occurring after knockdown of NKX2-2 (r = 0.67, p = 10³²) (Figure 3C). The results from these gene knockdown experiments therefore indicate that selected islet-specific lncRNAs and TFs can regu- late common gene expression programs.

Islet TFs and lncRNAs Co-regulate Genes Associated with Enhancer Clusters

Recent studies have revealed that islet TFs regulate cell-specific transcription by targeting clusters of enhancers and, in partic- ular, clusters with enhancers that are bound by multiple islet TFs (Pasquali et al., 2014). Enhancer clusters share many fea- tures with regulatory domains that have otherwise been defined as ‘‘stretch enhancers’’ or ‘‘superenhancers’’ (Pasquali et al., 2014; Pott and Lieb, 2015). Given that knockdown of islet lncRNAs and TFs suggested that they regulate similar genes, we asked whether islet lncRNAs also regulate enhancer clus- ter-associated genes. As expected, gene set enrichment anal- ysis (GSEA) showed that genes with islet-enriched expression, genes associated with enhancer clusters, or genes associated with enhancers that are bound by multiple TFs were downregu- lated after knockdown of all five TFs, whereas this was not observed for ten control sets of genes expressed at similar levels (Figure 4;Figures S4A and S4B). Likewise, genes associated with enhancer clusters and those showing islet-specific expres- sion were also enriched among genes that were downregulated after knockdown ofHI-LNC12,15,30,78,80, 85, and71(Fig- ure 4;Figures S4A and S4B). These results therefore indicate that islet-specific TFs and lncRNAs often co-regulate genes that are associated with enhancer clusters.

bCell lncRNAs and TFs Form Part of Islet-Specific Co-expression Networks

We next used an independent experimental approach to validate the observation that humanbcell lncRNAs and TFs regulate common gene expression programs. This involved the analysis of gene modules that show co-expression across a panel of hu- man islet RNA samples. Analogous approaches have been em- ployed to reveal sets of genes that share functional relationships (Derry et al., 2010; Kim et al., 2001; Pandey et al., 2010; Segal et al., 2003; Stuart et al., 2003; Su et al., 2011). We implemented this analysis using weighted gene co-expression analysis (WGCNA) of RNA sequencing (RNA-seq) profiles from 64 human pancreatic islet samples. This identified 25 major gene modules containing more than 100 genes, named M1–M25, that showed highly significant co-expression across human islet samples (Figure 5A;Table S2). We next determined which co-expression modules contained islet lncRNAs. Rather than using our previ- ously defined set of lncRNAs, this analysis was performed with a set of 2,373bcell lncRNAs that was newly annotated using 5 billion stranded RNA-seq reads pooled from 41 islet samples (Table S3;Figure S5A).bCell lncRNAs were found to be enriched in seven pancreatic islet co-expression modules (M3, M7, M12, M13, M18, M20, and M21) (Figure 5B).

We next characterized the nature of these seven lncRNA-en- riched co-expression modules. Five of these (M3, M7, M12, M18, and M20) were enriched in genes associated with pancre-

A B

NKX2.2 KD (Log2 FC)

Control KD (Log2 FC)

HI-LNC15 KD (Log2 FC)

r=0.67

p=10^-32 r=-0.1 p=0.08 HI-LNC78 KD

(Log2 FC)

HNF1A KD

(Log2 FC) MAFB KD (Log2 FC)

Control KD (Log2 FC)

r=0.89 p=10^-79

r=0.87 p=10^-71

r=-0.15 p=0.02

0 0.5 1

PDX1 HI-LNC15 HI-LNC78 HI-LNC80

GLIS3 HNF1AMAFB

NKX2.2 HI-LNC12 HI-LNC71 Control_1 Control_3 Control_5 Control_6 Control_7 Control_8 Control_2 Control_9 Control_4 Control_10

Downregulated Upregulated

C

HI-LNC12 HI-LNC15 HI-LNC30 HI-LNC75 HI-LNC76 HI-LNC78 HI-LNC80

GLIS3 HNF1A MAFB NKX2.2 PDX1 HI-LNC25 HI-LNC70 HI-LNC71 HI-LNC79 HI-LNC85

HI-LNC12 HI-LNC15 HI-LNC30 HI-LNC75 HI-LNC76 HI-LNC78 HI-LNC80 GLIS3 HNF1A MAFB NKX2.2 PDX1 HI-LNC25

HI-LNC79 HI-LNC71

HI-LNC85 HI-LNC70

Figure 3. Human Islet TFs and lncRNAs Regulate Common Genes

(A) Heatmap displaying Pearson r values for all pairwise comparisons of fold changes in gene expression after knockdown of TFs and lncRNAs.

Only genes significantly dysregulated at p < 10³ under at least one condition were included in the analysis.

(B) Unsupervised clustering analysis of fold change values after knockdown of five TFs and the five lncRNAs that displayed the strongest transcrip- tional changes. Only genes that were dysregulated at p <10³ in at least one knockdown were selected. Blue represents downregulated and red represents upregulated genes. Controls represent control comparisons as described forFigure 1.

(C) Examples of highly correlated transcriptional phenotypes. The plots show fold change values (Log2) after knockdown of the indicated pairs of genes. Only the top 100 most regulated genes for any of the two knockdowns were plotted. Pear- son’s correlation (r) and p values are displayed.

atic islet enhancer clusters (Figures 5A–

5C, marked in blue). Two other modules (M13 and M21) were enriched for ubiquitously expressed genes involved in mRNA translation and metabolic pathways (Figure S5B). Among the modules enriched in lncRNAs and enhancer clusters, three (M3, M7, and M18) were also enriched in islet-specific TF genes (Figure 5D), and two of these modules (M3 and M7) con- tained nine of the 12 lncRNAs that had been knocked down in EndoC-bH1 cells. Module M3, the largest of the seven lncRNA-enriched modules, featured gene ontology (GO) terms associated with prototypical islet cell functions and contained several islet TFs and lncRNAs (Figure 5E). In keeping with these findings, we found numerous instances of islet lncRNAs and known cell-specific TFs that showed a tight correla- tion of gene expression levels across human islet samples (Fig- ure 5F;Figure S5C). These findings thus indicated thatbcell-spe- cific lncRNAs, TFs, and genes associated with islet enhancer clusters form part of common expression programs.

Further analysis is consistent with the notion that lncRNAs play a functional role in driving gene expression variation in the lncRNA-enriched co-expression modules. First, the subset of lncRNAs that were shown to regulate an adjacent gene in knock- down studies also exhibited a particular high co-regulation with the adjacent gene across islet samples (Figure S1G). This obser- vation was extended to define 292 lncRNAs that displayed a highly significant (p < 10⁷) correlation of expression with an adjacent protein-coding gene in the panel of human islet sam- ples and are thus candidatecis-regulatory lncRNAs (Table S6).

Second, we analyzed all genes that were significantly downregu- lated in EndoC-bH1 cells after knocking downHI-LNC12,71,78, and80and found that they were also enriched among genes in human islet modules M3, M7, and M18 but not in size-controlled modules (Figure S5D). In summary, co-expression analysis of native human islets corroborated the findings observed with 404 Cell Metabolism25, 400–411, February 7, 2017

amiRNA-based perturbations in EndoC-bH1 cells and indicated that a group of islet lncRNAs and TFs form part of common tran- scriptional networks that target clusters of pancreatic islet en- hancers (Figure 5G).

Deregulation ofbCell lncRNAs in Human T2D

The identification of functional lncRNAs led us to explore whether some lncRNAs are abnormally expressed in human T2D and might thus be relevant to the pathogenesis of this dis- ease. We therefore analyzed our new set of 2,373 lncRNAs in a recently reported gene expression dataset that includes human islet samples from donors diagnosed with T2D or impaired glucose tolerance (IGT) (Fadista et al., 2014). Our results showed that, despite the fact that gene expression across human islet donors is highly variable, the expression of 15 and 100 lncRNAs was significantly altered in islets from T2D and IGT versus non- diabetic donors respectively (adjusted p < 0.05) (Figure S6A;

seeTable S7for a complete list). This finding suggests a poten- tial role of functionalbcell lncRNAs in driving some of thebcell gene expression changes that are associated with T2D.

PLUTORegulatesPDX1, an Essential Transcriptional Regulator

To explore howbcell lncRNAs can regulate cell-specific tran- scriptional networks, we focused onHI-LNC71, a nuclearly en- riched transcript (Figure S1B) that is transcribed from a promoter that is located3 kb upstream ofPDX1, in an antisense orienta- tion (Figure S6B). PDX1 is an essential transcriptional regulator of pancreas development andbcell function that has been impli- cated in genetic mechanisms underlying Mendelian and type 2 diabetes (Ahlgren et al., 1998; Jonsson et al., 1994; Offield et al., 1996; Stoffers et al., 1997). Based on this genomic location, we re- namedHI-LNC71 PLUTO, forPDX1locus upstream transcript.

The potential importance ofPLUTOwas strengthened by the observation thatPLUTOwas among the most markedly downre- gulated lncRNAs in islets from T2D or IGT donors (adjusted p value = 0.07 and 0.005, respectively;Figure 6A;Figure S6B).

Interestingly, PDX1 was also downregulated in islets from donors with T2D and IGT (Figure 6A).

PLUTOis a multi-isoform transcript that contains five major exons that span nearly 100 kb, encompassing a cluster of en- hancers that make 3D contacts with thePDX1promoter in hu- man islets and in EndoC-bH1 cells (Figure 6B; Figure S6A).

p<0.05 0

1 2 3

GSEA FWER p-value (-Log10)

HI-LNC12 HI-LNC15 HI-LNC30 HI-LNC75 HI-LNC76 HI-LNC78 HI-LNC80

GLIS3 HNF1A MAFB NKX2.2 PDX1

HI-LNC25 HI-LNC79 HI-LNC71 HI-LNC85

Ctl vs Ctl HI-LNC70

Enhancer cluster genes Control gene sets (n=10 )

Figure 4. LncRNAs Regulate Enhancer Cluster Genes

GSEA showed that genes that were down- regulated upon knockdown of either islet TFs or lncRNAs were enriched in a set of 694 genes that is associated with human islet enhancer clusters (red dots) but not in ten control gene sets (black dots) that were expressed at similar levels as enhancer cluster genes.

This observation suggested thatPLUTO could affect cis regulation of the PDX1 gene.

To test whether PLUTO regulates PDX1, we first examined EndoC-bH1 cells after amiRNA-mediated knockdown of PLUTORNA and found reducedPDX1mRNA and protein levels (Figure 6C). Simi- larly, knockdown ofPLUTORNA in dispersed primary human islet cells caused decreasedPDX1mRNA (Figure 6D). To vali- date these experiments through a complementary approach, we used CRISPR interference (CRISPRi), which involves target- ing guide RNAs (gRNAs) downstream of a gene’s transcriptional initiation site to block its transcription. Two independent gRNAs that targeted a region downstream of thePLUTOinitiation site efficiently reducedPLUTORNA levels relative to non-targeting gRNAs, and, in both cases, this led to decreasedPDX1mRNA expression (Figure 6E). Therefore, perturbing either PLUTO RNA levels or its transcription leads to the same inhibitory effect onPDX1mRNA.

The mousePdx1locus also has an islet lncRNA (Pluto) that shows only limited sequence homology with human PLUTO.

Plutois also transcribed from the opposite strand ofPdx1but is initiated from a promoter within the first intron ofPdx1and, like PLUTO, spans a broad regulatory domain upstream of Pdx1 (Figure S6C). Knockdown of Pluto RNA in the mouse bcell line MIN6 also led to decreasedPdx1mRNA levels (Fig- ure S6E). These experiments therefore indicated that PLUTO regulatesPDX1mRNA in human bcell lines and primary islet cells, and an analogous effect was observed for the mouse lncRNA ortholog.

Consistent with this regulatory relationship,PLUTOandPDX1 RNA levels are highly correlated across islet samples (Pearson’s r = 0.86, p = 10¹⁵;Figure 6F), and knockdown ofPDX1and PLUTO in EndoC-bH1 cells resulted in the deregulation of a shared set of genes (Figures 6G–6J). Furthermore, Plutoand Pdx1were found to be regulated with nearly identical dynamics in response to a shift in glucose concentration (4–11 mM) in mouse pancreatic islets (Figure S6D).PLUTOandPDX1there- fore regulate a common program in pancreatic islets, and this is at least in part explained by the fact that PLUTO regulatesPDX1.

PLUTORegulatesPDX1Transcription and Local 3D Chromatin Structure

To assess the mechanisms underlying the function ofPLUTO, we first examined whetherPLUTOcontrols the stability or tran- scription ofPDX1.Transcriptional inhibition experiments using Actinomycin D showed no significant differences in the stability

of PDX1 mRNA upon PLUTO knockdown (Figure 7A). By contrast, intronicPDX1RNA was reduced uponPLUTOknock- down, suggesting that PLUTO regulates PDX1 transcription (Figure 7B).

BecausePLUTOspans an enhancer cluster, we hypothesized that it could regulate the chromatin state of active enhancers. We thus knocked downPLUTOinbcells and measured H3K27 acet- ylation as well as H3K4 mono- and tri-methylation levels at several enhancers within the cluster. Our results indicate no sig- nificant changes in these characteristic active chromatin marks (Figure S7).

We next determined whetherPLUTOaffects the 3D contacts between the enhancer cluster and thePDX1promoter. Examina- tion of thePDX1locus using quantitative chromatin conforma- tion capture (3C) assays revealed that two far upstream enhancers (Figure 7C) showed reduced contacts with the PDX1 promoter after PLUTO knockdown (Figure 7D). These

PDX1 NKX2.2 NKX6.1 MAFA MAFB PAX6 NEUROD1

A

Ion transport Synaptic transmission Glucose homeostasis Nervous system dev.

Secretion by cell

TFs and lncRNAs in M3

E D

C

Enrichment p-value (-log10) 0 20 40

M3 M18 M20 M7 M12 Islet enhancer

cluster genes

Islet specific TFs M3 M18 M7 M11

Enrichment p-value (-log10) 0 4 8

HI-LNC12 HI-LNC15 HI-LNC71 HI-LNC80

F _r=0.78

NKX6-1 mRNA

HI-LNC78 RNA

r=0.71

NKX6-1 mRNA

HI-LNC80 RNA

HI-LNC12 RNA HI-LNC15 RNA

PAX6 mRNA PAX6 mRNA

r=0.65 r=0.7

GAPDH mRNA

HI-LNC78 RNA r=-0.07

HI-LNC80 RNA

GAPDH mRNA

r=0.08

HI-LNC12 RNA GAPDH mRNAHI-LNC15 RNA

GAPDH mRNA

r=-0.07 r=-0.15

M3: Gene ontology terms

0 6 12 Enrichment p-value (-log10)

B

G

M7 M3

30 Enrichment p-value (-log10) 0 20

Islet lncRNAs

10

M13 M20 M12 M18 M21 M3

M7 M18 M20 M21 M12

M13

Figure 5. Islet-Specific Coding and Noncoding RNAs Form Shared Co-expression Modules (A) Topological overlap matrix representing co- expression modules that were co-regulated across 64 human islet samples. Modules that were enriched in lncRNAs are marked with squares (hypergeometric test, p < 10²).

(B–D) Co-expression modules that showed enrichment in islet lncRNAs (B), islet enhancer cluster (EC)-associated genes (C), or a set of 94 islet-enriched TF genes (D). Five modules (M3, M7, M12, M18, and M20, marked in blue) out of seven modules that were enriched in lncRNAs were also enriched in ECs and TFs.

(E) Module M3 was enriched in typical islet-specific biological process annotations. Right: examples of islet TFs and lncRNAs in module M3.

(F) Correlation of the indicated lncRNAs andbcell- specific TF mRNAs across 64 islet samples.GAPDH is shown as a non-bcell reference. Pearson’s corre- lation values are displayed in the top left corner. The axes show expression values normalized across 64 islet samples.

(G) Network diagram illustrating that TFs and lncRNAs often co-regulate the same genes, many of which are associated with enhancer clusters.

findings therefore show thatPLUTO regu- lates the transcription of PDX1, a key pancreaticb cell transcriptional regulator, and that this is associated with its ability to promote contacts between the PDX1 promoter and its enhancer cluster (Figure 7E).

DISCUSSION

In the current study, we have tested the hypothesis that lncRNAs play a role in cell-specific gene regulation in pancreatic bcells, a cell type that is central in the path- ogenesis of human diabetes. We have thus carried out, for the first time, a system- atic analysis of the function of a set of human b cell-specific lncRNAs. Our experiments revealed several examples ofbcell lncRNAs in which sequence-specific perturbation causes transcriptional and functional phenotypes.

We have further shown that b cell-specific lncRNAs and TFs regulate a common transcriptional network. Finally, we have demonstrated thatbcell-specific lncRNAs directly or indirectly participate in the regulation of human enhancer clusters, which are the major functional targets of islet-specific transcription factors and key cis-regulatory determinants of islet cell tran- scriptional programs (Pasquali et al., 2014). Importantly, these conclusions are supported by concordant results from co- expression network analysis and loss of function experiments.

These studies should be interpreted in light of previous evi- dence indicating that a significant fraction of lncRNAs show lineage-specific expression (Cabili et al., 2011; Derrien et al., 2012; Goff et al., 2015; Guttman et al., 2011; Iyer et al., 2015;

Mora´n et al., 2012; Pauli et al., 2012). Our study extends 406 Cell Metabolism25, 400–411, February 7, 2017

previous findings by demonstrating a functional role of lncRNAs in lineage-specific TF networks.

Our findings invite the question of what molecular mecha- nisms underlie the regulatory effects ofbcell lncRNAs. LncRNAs have been proposed to control gene expression through diverse molecular mechanisms, including the formation of protein-spe- cific interactions and scaffolds, RNA-DNA or RNA-RNA hybrids, the titration of miRNAs, and the modulation of 3D chromosomal

structures (Rinn and Chang, 2012; Wang and Chang, 2011), whereas some transcripts currently defined as lncRNAs can theoretically encode for atypical small peptide sequences (Andrews and Rothnagel, 2014). Our knockdown and co- expression analyses have identified a subset of functional lncRNAs that appear to regulate a nearby gene, suggesting a lncRNA-based cis-regulatory mechanism, whereas others are likely to exert trans-regulatory effects. We focused on one

A B

C

D E

F G H

I J

Figure 6. PLUTO Knockdown Decreases PDX1mRNA

(A) Downregulation of PLUTO (HI-LNC71) and PDX1in islets from donors with T2D or IGT. Dif- ferential expression analysis was performed on control (n = 50) versus T2D (n = 10) or IGT (n = 15) samples. Boxplots represent expression normal- ized to the mean of control samples. Adjusted p values are shown.

(B) Schematic of the humanPDX1locus and its associated enhancer cluster. A 4C-seq analysis was designed to identify regions interacting with thePDX1 promoter region in EndoC-bH1 cells.

Red and orange vertical lines depict active and poised islet enhancers, respectively. F and R represent forward and reverse RNA-seq strands, respectively, and scales represent RPM.PLUTO (HI-LNC71) was generated from a de novo as- sembly of islet RNA-seq and differs from a tran- script annotated in UCSC and RefSeq that origi- nates from aPDX1intronic region.

(C) Downregulation of PLUTO or PDX1 using amiRNAs resulted in reducedPDX1mRNA and protein levels. EndoC-bH1 cells were transduced with control (black),PLUTO(white), orPDX1(tur- quoise) amiRNA vectors 80 hr prior to harvest.

RNA levels were assessed by qPCR, normalized to TBP, and expressed as fold over control amiRNA samples (n = 4). For protein quantification, PDX1 levels were first normalized to the average of TBP and H3 levels and then compared with the control amiRNA sample.

(D) Downregulation ofPLUTOin human islet cells results in reducedPDX1mRNA levels. Islet cells were dispersed and transduced with amiRNA vectors (n = 3) as in (B).

(E) Downregulation ofPLUTOin EndoC-bH3 cells using CRISPRi also decreasesPDX1mRNA. En- doC-bH3 cells were nucleofected with CRISPRi vectors 80 hr prior to harvest. RNA levels were assessed by qPCR and normalized toTBPand then to a control CRISPRi sample (n = 3).

(F) PDX1 and PLUTO RNA levels were highly correlated in 64 human islet samples.

(G) Knockdown ofPDX1andPLUTOresulted in differential expression of similar genes. Fold change value (Log2) of top 250 dysregulated genes following thePDX1knockdown was plotted against the same genes following the PLUTO knockdown.

(H) GSEA showed that genes that were down- regulated upon knockdown ofPDX1andPLUTO were enriched in genes whose enhancers were bound by PDX1 (red) in islets but not in ten control gene sets (black) that were expressed at similar levels as PDX1-bound genes.

(I) Knockdown ofPDX1andPLUTOresulted in differential expression of genes with similar biological process annotations.

(J) Examples of knownPDX1-regulated genes that are also co-regulated byPLUTOin parallel knockdown experiments. mRNA levels were assessed as in (B).

Error bars denote SEM; ***p < 10³, **p < 0.01, *p < 0.05 (Student’s t test).

functional nuclear-enrichedbcell lncRNA,PLUTO, and found that its function inbcell networks is at least in part due to its ability to elicit an effect on the transcription of its adjacent gene,PDX1, which encodes a keyb cell transcription factor.

Importantly, this was observed for both the mouse and human orthologs, and similar effects were obtained through RNAi suppression or through CRISPR-induced transcriptional interfer- ence ofPLUTO. Our studies further showed thatPLUTOpro- motes 3D interactions between the PDX1 promoter and its upstream enhancer cluster, which is contained within the body of thePLUTO gene. We thus propose that PLUTO regulates the 3D architecture of the enhancer cluster at thePDX1locus.

This finding is reminiscent but distinct from earlier examples of non-coding RNA genes that modulate 3D chromosomal struc- ture (Lai et al., 2013; Yao et al., 2010). Given that a significant number of lncRNAs are co-expressed with adjacent lineage- specific protein-coding genes, it is possible that the general regulatory paradigm described here is relevant to analogous lncRNA-protein coding gene pairs.

Taken together, our data implicate cell-specific lncRNAs in humanb cell transcriptional programs. Given the importance of TFs in the pathophysiology of human diabetes and their role inbcell programming strategies, it now seems reasonable to explore whether bcell lncRNAs also play analogous roles

A B

C

D E

Figure 7. PLUTORegulatesPDX1Transcrip- tion and 3D Chromatin Structure

(A) The mRNA stability ofPDX1was unaffected by PLUTOknockdown.PDX1mRNA was measured in control andPLUTOamiRNA knockdown in EndoC- bH1 cells after Actinomycin D (ActD) treatment (n = 3). mRNA levels are presented as a percentage of levels observed at time = 0.

(B) Knockdown ofPLUTOwas carried out as in Figure 6B, and this led to reduced PDX1 tran- scription, as assessed by qPCR analysis of intronic PDX1RNA levels using hydrolysis probes. Values were normalized toTBPmRNA and expressed as fold over the control amiRNA sample (n = 4).

(C) Schematic of selected epigenomic features of thePDX1locus.

(D)PLUTOis required for 3D contacts between the PDX1promoter and distal enhancers. 3C analysis revealed that knockdown ofPLUTO resulted in reduced contacts between the PDX1 promoter (anchor) and two enhancers (E1 and E2). Interac- tion signals were normalized to a control region on thePDX1intron. CTL represents a negative control region that does not harbor interactions with the PDX1promoter. Error bars denote ± SEM, and p values are from a Student’s t test.

(E)PLUTO knockdown resulted in impaired 3D contacts between the PDX1 promoter and its adjacent enhancer cluster, causing reducedPDX1 transcriptional activity.

(Bell and Polonsky, 2001; Flanagan et al., 2014; Zhou et al., 2008). The findings re- ported here therefore strengthen earlier suggestions that defects inbcell lncRNAs might contribute to the pathogenesis of human diabetes (Fadista et al., 2014;

Mora´n et al., 2012) and warrant an assessment of whether they can be harnessed to promotebcell differentiation, function, or cellular mass.

EXPERIMENTAL PROCEDURES Pancreatic Islets

Human islets used for RNA-seq and chromatin immunoprecipitation sequencing (ChIP-seq) were cultured with CMRL 1066 medium containing 10% fetal calf serum (FCS) before shipment, after which they were cultured for 3 days with RPMI 1640 medium containing 11 mM glucose and supple- mented with 10% FCS.

Glucose-Stimulated Insulin Release

Glucose-stimulated insulin release was assayed in EndoC-bH1 or EndoC-bH3 cells as described previously (Benazra et al., 2015; Ravassard et al., 2011).

RNA Analysis

RNA was isolated with Tripure (Roche) and treated with DNase I (Sigma). qPCR was performed with SYBR green or Taqman probe detection (van Arensber- gen et al., 2010). SeeTable S4for oligonucleotide and probe sequences.

amiRNA and CRISPRi Experiments

Lentiviral vectors carrying amiRNAs targeting TFs, lncRNAs, and non-targeting control sequences were transduced into the EndoC-bH1 humanbcell line as described previously (Castaing et al., 2005; Ravassard et al., 2011; Scharf- mann et al., 2014).

408 Cell Metabolism25, 400–411, February 7, 2017

Figure S1A illustrates the vector design. Oligonucleotide sequences are shown inTable S4. Non-transduced cells were assayed in parallel. Cells were harvested 80 hr post transduction for RNA extraction. For transduction of human islets, islets were first dispersed using trypsin-EDTA and gentle agitation. CRISPRi experiments were performed with two gRNAs designed to targetPLUTOexon 1 or two unrelated intergenic control regions and trans- fected in EndoC-bH3 cells (Table S4).

Gene Expression Array Analysis

RNA was hybridized onto HTA2.0 Affymetrix arrays. RMA normalization was carried out using Expression Console (Affymetrix). Gene-based differential expression analysis was done using Transcriptome Analysis Console (TAC, Affymetrix). Enhancer cluster genes were defined by genes that were associated with clustered islet enhancers that show top 50 percentile binding by TFs (PDX1, FOXA2, NKX2-2, NKX6.1, and MAFB) as defined previously (Pasquali et al., 2014). Pancreatic islet gene sets used for enrichment analysis are shown inTable S5. A list of islet- enriched genes was generated as those with more than two SDs higher expression in human islets than the average expression in 16 human tissues (Table S5).

Differential Expression in IGT and T2D Islets

RNA-seq data have been described previously (Fadista et al., 2014). The sam- ples were aligned to the hg19 genome using STAR aligner version 2.3.0 as described in theSupplemental Experimental Procedures, quantification was carried out with HTseq-Count 0.6.1, and differential expression analysis of lncRNA genes was done using DEseq2 1.10 (Table S3) using an adjusted p value threshold of 0.05.

3C

3C and 4C-seq was carried out as described previously (Pasquali et al., 2014;

Tena et al., 2011) For real-time PCR quantification, readings were normalized to a control region within thePDX1intron. Normalized values are expressed as a fraction of non-targeting amiRNA control sample. SeeTable S4for oligonu- cleotide sequences.

Annotation of Islet lncRNAs

LncRNAs were annotated through de novo assembly of5 billion stranded paired-end RNA-seq reads from 41 human islet samples, filtered for expres- sion in FACS-purifiedbcell cells, lack of enrichment in the pancreatic exocrine fraction to exclude acinar contaminants, and the presence of H3K4me3 enrichment in the vicinity of the 5⁰end. A more detailed description of the anno- tation process is provided in the Supplemental Information. Annotations are available inTable S3and can be accessed on a UCSC Genome Browser (GRCh37/hg19) session by selecting ‘‘track hubs’’ and ‘‘Human Islet lncRNAs.’’ Alternatively, the track hub can be directly visualized in the UCSC Genome Browser using the following link:http://genome.ucsc.edu/cgi-bin/

hgTracks?db=hg19&hubUrl=http://www.imperial.ac.uk/medicine/beta-cell- genome-regulation-laboratory/data/HILNCs/HILNCs.txt&hgS_loadUrlName=

http://www.imperial.ac.uk/medicine/beta-cell-genome-regulation-laboratory/

data/.

Network Analysis

The WGCNA(v2) tool was used to build a co-transcriptional network based on mRNAs from 64 human islet RNA-seq samples.

ACCESSION NUMBERS

The accession number for the microarray data reported in this paper is GEO:

GSE83619.

SUPPLEMENTAL INFORMATION

Supplemental Information includes Supplemental Experimental Procedures, seven figures, and seven tables and can be found with this article online at http://dx.doi.org/10.1016/j.cmet.2016.11.016.

AUTHOR CONTRIBUTIONS

J.F. and I.A. conceived the idea, designed the experiments, and wrote the manuscript. J.F. supervised and I.A. coordinated the project. I.A., Z.T., H.W., J.Y., C.A., E.S., A.S., L. Pasquali, and D.M.Y.R. contributed to data analysis.

I.A., A.B., M.B., C.S.C., R.G.F., J.G.H., and N.C. performed the experiments.

D.M.Y.R., I.M., and N.N. annotated lncRNAs. L. Piemonti, T.B., C.B., J.K.C., and F.P. provided samples. I.A., J.F., Z.T., A.B., D.M.Y.R, L.G., C.B., J.K.C., F.P., P.R., A.S., L.G., C.A., and E.S. discussed the results. All authors read and approved the manuscript. Z.T., A.B., D.M.Y.R., C.S.C., and N.N. contrib- uted equally.

ACKNOWLEDGMENTS

The research was supported by the National Institute for Health Research (NIHR) Imperial Biomedical Research Centre. Work was funded by grants from the Wellcome Trust (WT101033 to J.F.), NIH-BCBC (2U01 DK072473- 06 to J.F. and P.R.), Ministerio de Economı´a y Competitividad (BFU2014- 54284-R to J.F.), and Horizon 2020 (667191 to J.F.). Work in IDIBAPS was sup- ported by the CERCA Programme, Generalitat de Catalunya. J.Y. was sup- ported through a Berg and Unity Biotechnology fellowship. The authors are grateful to Helena Raurell Vila for experimental help and Romain Derelle and Loris Mularoni for advice regarding bioinformatics analysis. P.R. is a share- holder and consultant for Endocells/Unicercell Biosolutions. Z.T. receives financial support from Berg Pharma and Unity Biotechnology as a consultant.

Received: May 5, 2016 Revised: October 1, 2016 Accepted: November 29, 2016 Published: December 29, 2016 REFERENCES

Ahlgren, U., Jonsson, J., Jonsson, L., Simu, K., and Edlund, H. (1998). beta- cell-specific inactivation of the mouse Ipf1/Pdx1 gene results in loss of the beta-cell phenotype and maturity onset diabetes. Genes Dev.12, 1763–1768.

Anderson, K.R., Singer, R.A., Balderes, D.A., Hernandez-Lagunas, L., Johnson, C.W., Artinger, K.B., and Sussel, L. (2011). The L6 domain tetraspa- nin Tm4sf4 regulates endocrine pancreas differentiation and directed cell migration. Development138, 3213–3224.

Andrews, S.J., and Rothnagel, J.A. (2014). Emerging evidence for functional peptides encoded by short open reading frames. Nat. Rev. Genet. 15, 193–204.

Bell, G.I., and Polonsky, K.S. (2001). Diabetes mellitus and genetically pro- grammed defects in beta-cell function. Nature414, 788–791.

Benazra, M., Lecomte, M.J., Colace, C., M€uller, A., Machado, C., Pechberty, S., Bricout-Neveu, E., Grenier-Godard, M., Solimena, M., Scharfmann, R., et al. (2015). A human beta cell line with drug inducible excision of immortal- izing transgenes. Mol. Metab.4, 916–925.

Benner, C., van der Meulen, T., Cace´res, E., Tigyi, K., Donaldson, C.J., and Huising, M.O. (2014). The transcriptional landscape of mouse beta cells compared to human beta cells reveals notable species differences in long non-coding RNA and protein-coding gene expression. BMC Genomics 15, 620.

Cabili, M.N., Trapnell, C., Goff, L., Koziol, M., Tazon-Vega, B., Regev, A., and Rinn, J.L. (2011). Integrative annotation of human large intergenic noncoding RNAs reveals global properties and specific subclasses. Genes Dev.25, 1915–1927.

Carninci, P., Kasukawa, T., Katayama, S., Gough, J., Frith, M.C., Maeda, N., Oyama, R., Ravasi, T., Lenhard, B., Wells, C., et al.; FANTOM Consortium;

RIKEN Genome Exploration Research Group and Genome Science Group (Genome Network Project Core Group) (2005). The transcriptional landscape of the mammalian genome. Science309, 1559–1563.

Carrieri, C., Cimatti, L., Biagioli, M., Beugnet, A., Zucchelli, S., Fedele, S., Pesce, E., Ferrer, I., Collavin, L., Santoro, C., et al. (2012). Long non-coding